1

ROE and Risk-Taking in Banks

December 2015

Christophe Moussu1

ESCP Europe, Labex Refi

Arthur Petit-Romec2

ESCP Europe, Labex Refi

JEL Classification: G01, G21, G28, G32

Keywords: Return on equity, risk-taking, financial crisis, bank performance

1Corresponding author: ESCP Europe, 79 avenue de la République 75543 Paris Cedex 11, France. Tel:+33 1 49 23 22 69; Fax: +33 1 49 23 20 80; E-mail: moussu@escpeurope.eu

2 ESCP Europe, 79 avenue de la République 75543 Paris Cedex 11, France. Tel: +33 1 49 23 20 33; Fax:+33 1 49 23 20 80; E-mail: arthur.romec@edu.escpeurope.eu

Acknowledgements: we would like to thank Jeff Pontiff and Steve Ohana for their important inputs on thepaper. We also thank Franck Bancel, Alexandre Garel, Cécile Kharoubi, Marti Subrahmanyan, MichaelTroege and Lei Zhao for helpful discussions and suggestions; as well as participants at the FEBSConference in Paris (2013), the Surrey Fordham Conference (2013), the MFA Conference in Orlando(2014), the CIG Conference in Dijon (2014), the AFFI Conference in Aix-en-Provence (2014) and theIFABS Conference in Lisbon (2014). Finally, we also thank seminar participants at the EIFR researchworkshop (2013), Audencia (2014), Cefag (2014), Labex Refi (2013) and at the BNPP- ESCP EuropeChair Cost of Capital workshop.

2

ROE and Risk-Taking in Banks

Abstract:

Return on equity (ROE) is a central measure of performance in the banking industry.

Empirical evidence that bank managers’ pay is linked to ROE exists. We investigate the

implications of ROE targeting on bank risk-taking. Using the 2007-2008 financial crisis, we

reveal that pre-crisis ROE has a strong and very robust positive impact on both individual and

systemic risk in the crisis. Our results are unchanged for the 1998 crisis but are totally reversed

when samples of non-banking firms are considered. ROE also explains the persistence of poor

performance across crises and appears as a key component of bank risk culture and vulnerability

to crises. Overall, our results strongly suggest that reliance on ROE as a performance measure is

a major incentive to take excessive risk in banks.

3

J. Dimon (CEO JP Morgan): “We feel good about the 17% ROE of the investment bank,” Financial

Times, January 16th, 2013

H. Schwartz (CFO Goldman Sachs): “There is a lot of value to having the best ROE in the industry,”

Financial Times, April 17th, 2013

1. Introduction

A great deal of anecdotal evidence suggests that return on equity (ROE) is a central measure

of performance in banks. Target ROEs are not only set at the bank level, but they are also used to

drive resource allocation across and even inside divisions. In a recent paper, Bennett, Gopalan

and Thakor (2015) document empirically that bank CEOs and executives have a larger portion of

their compensation linked to ROE than CEOs and executives of other industries. In this paper, we

analyze the implications of ROE targeting on bank risk-taking. Our main hypothesis is that ROE

is associated with higher risk and therefore carries strong risk-taking incentives in banks.

The reliance on ROE in banks has emerged from the risk management approach that inspired

bank capital regulation. In this approach, a specific “capital charge” is determined, depending on

the risk of each asset, in order to cover losses. This capital charge is either imposed by the

regulator or estimated through internal risk models. Therefore, looking for the highest return for a

given capital charge becomes the target, both inside divisions and at bank level. In this business

model of banks, bank capital is adjusted to risk and excess capital inefficient. This perspective is

hard to reconcile with a more traditional corporate finance approach, where risk is taken into

account through specific discount rates and “excess” capital does not matter, as it reduces both

the return and the risk for shareholders (Modigliani and Miller, 1958). Addressing the issue of

4

convergence of those approaches is not the purpose of this paper. However, given the prevalence

of ROE everywhere in banks, it is very important to provide an assessment of the incentives ROE

conveys in banks.

As Rajan (2006) argues, evaluating the true nature of bank performance is a very complicated

task, and risk-taking may seem value-enhancing, albeit as long as risk has not materialized. We

use the financial crisis as the materialization of the real risks taken by banks.

In order to test our main hypothesis, we first examine whether bank pre-crisis ROE is associated

with the realization of bank risk in the crisis. We use two measures of bank risk: bank stock

performance, as a proxy of bank individual risk, and the marginal expected shortfall (MES),

computed following Acharya et al. (2010), as a proxy of systemic risk. Our results reveal that

ROE is strongly associated with individual and systemic risk, based on a sample of 273 large

banks from 28 countries. Our results are very robust to the introduction of numerous variables,

controlling for risk, both on the asset and the liability sides of bank balance sheets, as well as

market measures of risk and country variables. In particular, our results hold when we introduce

control variables for bank capital or funding strategies, thereby indicating that ROE captures risk-

taking above and beyond funding issues. The results also hold when we restrict the sample to

deposit-taking banks. Interestingly, and contrarily to other control variables, ROE always appears

as a main determinant of both bank individual risk and systemic risk.

We analyze further the relationship between ROE and bank individual risk during the crisis,

by using dummy variables based on ROE quartiles. When tested against the lowest quartile, the

dummies for the other three quartiles are highly significant, which reveals that the association

between ROE and risk is not limited to banks with the highest ROE, but also holds for banks with

intermediate ROE. We also run similar tests for the 1998 crisis on a subsample of 181 banks and

5

uncover a similar relationship between bank pre-crisis ROE and individual risk in the 1998 crisis,

which suggests that the effect of ROE was not limited to the 2007-2008 crisis.

Three main reasons may explain our results, the more direct of which is a contingent

explanation. As argued previously, ROE emerged as the natural performance measure of a bank

business model grounded on risk management. With risk correctly assessed and capital exactly

adjusted to risk, expected ROE does not increase with bank risk. However, in good states of

nature, when losses are below their expected level, realized ROE increases with the level of risk.

From this perspective, a contingent effect may explain our results, in that a higher ROE before

the crisis resulted in higher risk that materialized during the crisis. One could argue that this

argument is trivial. A higher level of performance is achieved through higher risk and results in

higher losses when conditions worsen.

If the contingent effect is trivial, it should apply to firms outside the banking industry.

Consequently, we test whether our results apply to non-financial firms as well, running two

different out-of-sample tests. First, using the same empirical design and period, we test the

impact of pre-crisis performance measures on stock returns in the crisis for a sample of 449 non-

financial firms. Shareholder losses are comparable to those incurred by bank shareholders. In

sharp contrast with our results for banks, firms with higher pre-crisis ROCE (return on capital

employed), ROA (return on assets) or ROE performed better during the crisis. One reason for this

difference is that the 2007-2008 crisis is a banking crisis. In order to address this concern, we

propose a second out-of-sample test, in which we use the same empirical design for a shock

specific to another industry. We focus on the stock performance of oil companies after the recent

collapse of oil prices, which began in June 2014. We find no effect of pre-crisis performance

measure on stock returns when controlling for capital structure, asset structure, size, risk and

6

institutional variables. Our two out-of-sample tests suggest that the contingent effect does not

apply outside the banking industry. Therefore, the explanation is not trivial. The results reveal a

specificity of banks: stronger performance makes them more vulnerable in crisis.

Two other explanations complement or substitute for the contingent explanation, namely

mistakes and incentives. Concerning mistakes, our paper is related to Krüger, Landier and

Thesmar (2014), who reveal how the use of a unique company-wide discount rate led to

overinvestment in the riskiest activities of diversified firms. They attribute this mistake to a poor

command of financial principles. In banks, inadequate capital charges are equivalent to using

wrong discount rates. If capital charges are underestimated, ROE is artificially inflated and

associated with higher risk for shareholders. The opposite is true as well, in that the

overestimation of capital charges leads to lower ROE and lower risk. The complexity of bank

activities, the imperfection of regulation (e.g., Vallascas and Hagendorff, 2013; Mariathasan and

Merrouche, 2014), the existence of tail risk and heterogeneity in risk management practices (e.g.,

Aebi, Sabato, and Schmid, 2012; Ellul and Yerramilli, 2013) are possible reasons why banks

make mistakes. Bank capital regulation accentuates these potential errors; indeed, the system of

regulatory risk-weights does not provide a continuous adjustment of bank capital to risk, for

instance the same capital charge can be required for a class of assets with risk heterogeneity (e.g.,

sovereign bonds). When regulation relies on internal models, we are back to the issue of

heterogeneous risk management practices.

Our previous explanations apply even when mistakes in capital charges do not distort the bank

capital allocation process. However, as a central performance measure, ROE is intrinsically tied

to the resource allocation process in banks. Our third explanation is based on incentives. When

ROE is targeted, managers are inclined to distort the capital allocation process along two lines.

7

First, ROE targeting creates an incentive for managers to engage strategically in regulatory

arbitrage activities. Investing in the riskiest assets of a given class, favoring asset classes with

undervalued capital charges and underestimating the real risks are examples of such activities.

Empirical evidence highlights the existence of such regulatory arbitrage activities (Acharya,

Schnabl, and Suarez, 2013; Acharya and Steffen, 2015; Mariathasan and Merrouche, 2014).

Second, the use of ROE in banks is prone to a time-inconsistency problem. Asset allocation in

banks is supposed to be based on expected ROE, but the performance of bank managers is

assessed on observable realized ROE. Consequently, if ROE increases with risk in good times, in

line with a contingent effect, managers have clear incentives to inflate this measure taking higher

risk. Third, managers may be inclined to develop tail risk strategies, not taken into account in

capital charges. The positive association between ROE and systemic risk may flow from this

hypothesis. Also, when banks overcome regulatory constraints, by choosing the same assets with

underestimated capital charges, regulatory arbitrage leads to systemic risk.

Anecdotal evidence indicates that ROE is a main target in the design of the compensation

structure of bank managers. As mentioned previously, Bennett, Gopalan and Thakor (2015)

confirm empirically that bank CEOs and executives have monetary incentives to maximize ROE

on a sample of U.S. banks over the period 2006-2012. We check the existence of pay sensitivity

to ROE for a portion of our sample of international banks, for which CEO compensation is

available on Boardex. We find a strong sensitivity of CEO compensation to ROE, suggesting that

monetary incentives to maximize ROE existed in the years leading up to the crisis. The effect of

ROE is robust to the introduction of risk and size control variables, which are documented as

being important determinants of bank CEO compensation (Cheng, Hong, and Scheinkman,

8

2014). This adds to the empirical evidence of Bennett, Gopalan and Thakor (2015), as it provides

first evidence that CEO pay is also linked to ROE for non U.S. banks.

One interesting question is whether ROE is responsible for explaining the persistence of poor

bank performance across crises, documented by Fahlenbrach, Prilmeier, and Stulz (2012) for U.S.

banks. Following their methodology, we first show that the persistence of poor performance

across crises also holds for our international sample. Second, we reveal that banks with high

ROE, either in 1997 or in 2006, were more likely to perform poorly in both crises, which

suggests that ROE targeting is an important component of what Fahlenbrach et al. (2012)

describe as a ‘risk culture’.

Our paper adds to the prolific literature investigating the determinants of bank performance

during the crisis (Beltratti and Stulz, 2012; Berger and Bouwman, 2013; Demirgüç-Kunt,

Detragiache, and Merrouche, 2013; Erkens, Hung, and Matos, 2012; Fahlenbrach and Stulz,

2011; Minton, Taillard, and Williamson, 2014). Beltratti and Stulz (2012) in particular provide a

comprehensive study of the influence of both firm-level and country-level pre-crisis

characteristics on bank performance during the crisis. Our paper complements these previous

studies by revealing the role played by pre-crisis performance measures in explaining the

vulnerability of banks in crises. In particular, the effect of ROE is robust to the introduction of

bank capital structure variables, which are important determinant of an institution’s ability to

withstand shocks.

Our paper is also related to the literature investigating the role of incentives in banks.

Motivated by the fact that executive incentives were publicly blamed for the near-collapse of

9

banks in the financial crisis 3 , a prolific body of literature has investigated the relationship

between compensation and risk-taking. Numerous studies have examined this question, and

overall, the consensus is that compensation and risk in banks are strongly associated (Chesney et

al., 2011; De Young et al., 2013; Cheng, Hong, and Scheinkman, 2014; Balachandran et al.,

2011; Fahlenbrach and Stulz, 2011; Mehran and Rosenberg, 2007; Bolton, Mehran, and Shapiro,

2010). Our results complement this literature and indicate that one channel through which the

design of CEO compensation induced risk-taking is sensitivity to ROE.

Finally, our paper illustrates perfectly the existence of perverse effects associated with the

design of performance measures in firms. In particular, the fascination with an “objective”

criterion is viewed as the main cause of distorted incentives in organizations (e.g. Kerr, 1975;

Gibbons, 1998). Baker (1992) and Baker, Jensen, and Murphy (1988) discuss how the

focalization on a wrong or inadequate performance measure is likely to induce counterproductive

effects or agency conflicts if employees overreact to or game the performance measure. In

particular, Baker, Jensen, and Murphy (1988) explain that one important criticism of a pay-for-

performance system is that it can be too effective and motivate people to focus only on the output

used in the measure. In line with this idea, our results are consistent with the fact that managers

achieve higher ROE taking excessive risk.

The rest of the paper is organized as follows. The next section describes the sample, empirical

methodology and the variables. In section 3 we present and interpret the empirical results, while

in section 4 we propose two out-of-sample checks based on non-financial firms. In section 5 we

focus on the sensitivity of CEO compensation to ROE and in section 6 we investigate the

existence of an “ROE culture” in banks. Section 7 concludes.

3 See Wall Street Journal, “Crazy Compensation and the crisis,” by A. Blinder

10

2. Sample construction, description of variables and summary statistics

This section provides information on the construction of the sample and presents the main

variables used in the empirical analysis and some descriptive statistics.

2.1 Sample construction

We obtain accounting data on balance sheets and income statements from Bankscope,

stock and index returns from Datastream and CEO compensation information from BoardEx. The

starting point for the construction of the sample comprises all banks that existed in Bankscope

and Datastream databases for the year 20064 . We first exclude banks which are not listed,

because our main dependent variable is the buy-and-hold stock return of banks during the crisis.

Second, we reduce the sample to banks from those countries for which Djankov, La Porta,

Lopez-de-Silanes, and Shleifer (2008) provided revised data on anti-director rights and for which

regulation indices based on Barth, Caprio, and Levine (2008) are available. Finally, we focus our

analysis on the largest banks, with assets over $10 billion just before the crisis at the end of 2006.

To enter our sample, we require that a bank have a full set of data for our key variables.

We obtain a final sample of 273 banks from 28 countries (see Appendix A). In some

empirical tests, we also focus on a subsample restricted to deposit-taking banks. Following

4 Similar to previous studies (e.g. Fahlenbrach and Stulz, 2011; Beltratti and Stulz, 2012; Erkens et al., 2012), wechoose the year 2006 to construct our sample, because this was the last complete year before the crisis for which dataon balance sheets and income statements are available.

11

Beltratti and Stulz (2012), we impose a threshold of 20% for the ratio of customer-deposits to

total assets. This subsample is composed of 245 banks.

2.2 Empirical Methodology

The main model we test is a simple OLS model, in which the main dependent variable is

the buy-and-hold stock return of a bank in the crisis and the main independent variable is pre-

crisis ROE. We control for various characteristics at the bank level. Most control variables are

measured as for 2006, and we also control for several country-level variables. A definition of

each variable is provided in Appendix B.

- Dependent variable:

We investigate the determinants of individual bank risk during the crisis, by using buy-and-

hold returns, computed with weekly stock returns over a period of 18 months from July 2007 to

the end of 2008. Of course, bank stocks continued to suffer in 2009 as a result of the crisis.

However, consistent with the existing literature (Aebi, Sabato, Schmid, 2012; Beltratti and Stulz,

2012; Erkens, Hung, and Matos, 2012; Fahlenbrach and Stulz, 2011), we restrict the period to

2007-2008 in order to avoid bias in our dependent variable, since stock returns in 2009 were

affected to a certain extent by government interventions and uncertainty about the possible

nationalizations of banks.

In alternative specifications, we use the systemic risk contribution of a bank, using the

marginal expected shortfall (MES) as a dependent variable. Following Acharya et al. (2010), we

compute the MES as the average stock return of a bank over the worst 5% of days for the MSCI

12

world index during the period July 2007-December 2008. The MES corresponds to the drop in

bank market capitalization when the market experiences extreme downward movements, and

represents a measure of systemic risk.

- Main independent variable:

Our main independent variable is ROE, defined as the ratio of pre-tax profit to equity. We

choose to use pre-tax ROE in order to smooth tax differences across countries.

- Control variables at the bank level:

We focus on control variables at the bank level which have been highlighted previously in

the literature as important determinants of bank stock performance during the crisis5.

We employ two measures of bank capital. The first measure is the Tier 1 ratio, which

corresponds to the main regulatory capital ratio and is defined as the ratio of Tier 1 capital to total

risk-weighted assets. Our second measure is the tangible equity ratio, defined as tangible

common equity divided by total assets less intangible assets. The tangible equity ratio thus offers

a more conservative approach to bank capital than the standard ratio of equity to assets6.

We also use two other measures related to the capital structure of banks. The first measure

is the ratio of total customer deposits to total assets. Contrary to funding received from financial

markets or intermediaries, which can rapidly become expensive or unavailable, deposits are less

subject to runs because they are insured and therefore represent a stable source of funding.

Empirically, Beltratti and Stulz (2012) and Demirgüç-Kunt, Detragiache, and Merrouche (2013)

5 See, for example, Beltratti and Stulz (2012), Demirgüç-Kunt, Detragiache and Merrouche (2013), Erkens, Hungand Matos (2012).6 In unreported tests, we find that our results are unchanged if we consider the equity ratio.

13

confirm the positive impact of deposits on bank performance during the crisis. Our second

measure is the funding fragility ratio, proposed by Demirgüç-Kunt and Huizinga (2010), which is

defined as the ratio of the sum of deposits and short-term borrowing, excluding customer

deposits, to total deposits and short-term borrowing7.

Concerning the nature of activities, we focus on the importance of non-traditional

activities, by using the ratio of non-interest income to total operating income. Non-interest

income includes net fee income, net commission income and net trading income. Previous studies

have documented an impact of the share of non-traditional activities on bank risk, notably Baele,

De Jonghe, and Vander Vennet (2007) for the beta of the stock and Stiroh (2004) for the Z-score.

Furthermore, Demirgüç-Kunt and Huizinga (2010) show that expanding the share of non-interest

activities can offer some risk diversification at very low levels, though this approach quickly

becomes very risky.

We also control for the pre-crisis stock return over the year 2006, which can be

considered as a proxy for certain risks rewarded by the market prior to the crisis but which

eventually turned out badly during the crisis. In other words, stock performance during the crisis

could represent a reversal of pre-crisis performance. Empirically, Beltratti and Stulz (2012) and

Fahlenbrach et al. (2012) find that banks that performed well in 2006 suffered poor returns in the

crisis.

In additional tests, we also use the ratio of risk-weighted assets to total assets, often called

“density”, as a risk measure of bank assets. Alternatively, density can be considered as a proxy

for the regulatory arbitrage strategies pursued by banks in order to economize equity (Beltratti

7 It is therefore inversely related to the importance of customer deposits in the short-term funding of banks, which isthe most stable source of short-term funding.

14

and Paladino, 2013). In this case, lower density would indicate a more aggressive optimization of

risk weights in order to “save” capital.

Finally, we control throughout our analysis for the size and beta. Size is defined as the

natural logarithm of total assets, and we compute the beta by estimating a market model of

weekly returns from 2004 to 2006. We use the MSCI World Index as the market portfolio and the

three-month T-bill as the risk-free rate.

- Control variables at the country level:

On top of our previous control variables, we also control for six institutional variables at the

country level which may influence risk-taking behavior and stock performance in the crisis. Two

variables, namely the anti-director right index from Djankov et al. (2008) and the average of the

voice, political stability, government effectiveness, regulatory quality, rule of law and corruption

indicators from Kaufmann, Kraay, and Mastruzzi (2008), are general governance variables. The

other four are specific to the banking industry and correspond to the regulation variables set out

by Caprio, Laeven, and Levine (2007). Appendix B provides a precise description of these

variables. All country variables are included at the same time in the regressions, and the main

results are unchanged when entering them separately.

2.3. Summary statistics

Table 1 provides summary statistics for our bank sample. The median and mean buy-and-

hold stock returns for the sample, respectively, are minus 40.7% and minus 36.4% from July

2007 to December 2008. In line with previous studies (e.g. Beltratti and Stulz, 2012; Fahlenbrach

15

et al., 2012), the standard deviation of these returns, 32%, is rather high. The median and mean

MES are minus 3.27% and minus 3.69%. Pre-crisis ROEs were high, with a median and a mean

of 18.49% and 20.12% respectively, and some heterogeneity exists across banks, since the

standard deviation of ROE is 11% and the highest ROE reaches more than 50%. By contrast, the

mean and median ROA, 1.41% and 1.24%, were much smaller. This discrepancy in the levels of

ROE and ROA is related to the leverage of banks. Tangible equity ratios are relatively low, with

a median and a mean of 5.76% and 6.25%, respectively, while the median and mean Tier 1 ratios

are 8.57% and 9.08%, respectively. All banks are above the regulatory minimum of 4%. The

portion of deposits in the financial structure is important, with an average deposit ratio of 57.2%.

However, the standard deviation, 27%, is high, as our sample includes both commercial and

investment banks. Similarly, the average funding fragility ratio is 24.4%, with a standard

deviation of almost 27%. The average ratio of risk-weighted assets to total assets, 0.6, is

equivalent to that reported by Beltratti and Paladino (2013). The median bank in our sample has

$38 billion in assets in 2006, and the median and mean equity betas are 0.98 and 1.02,

respectively.

3. Presentation and discussion of the results

3.1 ROE and individual risk in the 2007-2008 crisis

Table 2 presents ten regressions in which the dependent variable is the bank buy-and-hold

stock return (BHR) during the crisis. The results indicate strongly that higher pre-crisis ROE is

associated with more value destruction for bank shareholders during the crisis. The negative

16

impact of ROE on bank stock returns appears to be highly significant, both statistically and

economically. According to regression 1, a one standard deviation increase in pre-crisis ROE is

associated with an 11.45% (-1.035*11.06%) lower BHR, which corresponds to 35.4% of its

standard deviation. This finding confirms that while in risk management models expected ROE

amounts to a risk-adjusted measure, realized ROE is in contrast inflated through higher risk in

good states of nature.

The impact of pre-crisis ROE on individual risk during the crisis remains very strong,

even after taking into account our different control variables. In all the specifications, ROE has a

strong and economically significant impact on the losses borne by bank shareholders during the

crisis, with rather strong stability in the coefficients. For example, according to regression 7,

where we control for the tangible equity ratio, the deposit ratio, the stock return in 2006, the size,

the beta and our country variables, a one-standard deviation increase in pre-crisis ROE remains

associated with a 6.48% lower BHR, which corresponds to 20% of its standard-deviation.

The impact of pre-crisis ROE on value destruction is robust and even increases when we

replace the tangible equity ratio with the Tier 1 ratio (regressions 6 and 8). Consistent with

Beltratti and Stulz (2012) and Demirgüç-Kunt, Detragiache, and Merrouche (2013), we find that

banks with more capital, more deposits and less funding fragility performed better during the

crisis, while results on our other control variables indicate that large banks and those with a

higher involvement in non-interest activities experienced worse stock returns during the crisis.

The negative impact of size is in line with the idea that large banks are more complex and had

more opaque asset and funding structures which were uncovered during the crisis. Alternatively,

the negative impact of size could also partly result from the existence of a ‘too big to save’ effect

(Demirgüç-Kunt and Huizinga, 2013). The negative impact of non-interest income is consistent

with Stiroh (2004) and Demirgüç-Kunt and Huizinga (2010), showing that the share of non-

17

interest income-generating activities is associated with higher risk. Banks with higher stock

returns in 2006 tend to have lower stock returns in the crisis, although the effect is not strongly

significant. Finally, banks with a higher ratio of risk-weighted assets experienced worse stock

returns during the crisis, which indicates that although probably imperfect, the pre-crisis

regulatory assessment of bank risk had some predictive power to explain individual risk. In

contrast, the beta appears to have no predictive power.

In regressions 9 and 10, we restrict the sample to deposit-taking banks (for which

customer deposits amount to at least 20% of total assets). Coefficients on ROE are similar and

even of slightly higher economic significance in the case of ROE, thereby indicating that our

results are not driven by investment banks.

Overall, the results from Table 2 indicate that ROE is associated with higher risk for bank

shareholders. Of course, if we could observe all the ex-ante strategies aiming at increasing risk,

we would expect to have no impact of ROE; however, the robustness and strength of the

predictive power of ROE indicate that it is a good proxy for those strategies.

In Table 3, we analyze whether there are asymmetries in the relation between ROE and

bank individual risk during the crisis. We split banks into quartiles, based on their pre-crisis

ROE, and then create dummy variables for each group. ROE is replaced by the quartile indicator

variables, and the omitted group is the quartile 1 corresponding to the group of banks with the

lowest pre-crisis ROE. The results for our control variables remain the same except for Tier 1

ratio, which is still positive but no longer significant. In the six specifications, all the quartile

dummies are highly significant and the coefficient becomes more and more negative as we move

from the lowest ROE quartile to the highest ROE quartile. For example, according to regression

1, being in the 2nd, 3rd and 4th quartiles is associated with a decrease in BHR of 16.6%, 22.8% and

24.7%, respectively. Remarkably, the BHR difference between the first and second quartiles is

18

already very important. These results go against the idea that the relation between ROE and

individual risk is driven by banks with extreme ROE only.

3.2 ROE and systemic risk in the 2007-2008 crisis

In order to better gauge the nature of the risk associated with reliance on ROE, we replace

the BHR with the MES, which in turn provides a measure of systemic risk. The results, which are

reported in Table 4, strongly indicate that banks with higher pre-crisis ROE had more systemic

risk during the crisis. Indeed, pre-crisis ROE has a strong and negative impact on the MES in all

specifications, and this effect appears to be highly significant both statistically and economically.

For example, according to regression 1, a one-standard-deviation increase in ROE is associated

with a decline of 0.56% (-5.069*11.06%) in the MES, which represents 24.5% of its standard

deviation.

Results for the control variables indicate that bank capital and deposits, which were

important determinants of bank stock returns during the crisis, have no predictive power for

systemic risk. Similarly, regressions 3 to 6 reveal that funding fragility and non-interest income

have no impact on MES. The results for our control variables are consistent with Laeven,

Ratnovski, and Tong (2014) and Weiß, Bostandzic, and Neumann (2014), in that they show that

the determinants of individual risk and systemic risk tend to be different for the recent financial

crisis. On the contrary, size is an important determinant of systemic risk, and interestingly, the

beta becomes significant to some extent in explaining MES whereas it had no impact on

individual risk.

19

3.3 The crisis of 1998

In this subsection we are interested in whether our results on ROE are specific to the recent

financial crisis. We check whether our results hold for another crisis, namely that of 1998, which

was marked by Russia’s default and the collapse of long-term capital management (LTCM) and

was at that time considered the worst financial crisis since the Great Depression.

Table 5 presents regressions for which the dependent variable is the buy-and-hold stock

return of a bank during the crisis of 1998, while the ROE and control variables are computed for

the year 1997. We follow Fahlenbrach, Prilmeier, and Stulz (2012) and use daily return data to

compute buy-and-hold returns from August 3, 1998 (the first trading day of August 1998) to the

day on which the bank attains its lowest stock price across the rest of the year 1998. Table 5

indicates that the results on ROE are unchanged for this specific crisis and robust to the

introduction of several control variables. The effect appears to be highly significant, both

statistically and economically. The results for the control variables indicate that, as in 2007-2008,

large banks and banks with higher non-interest income experienced worse stock returns during

the 1998 crisis. Conversely, the deposit ratio and funding fragility have no predictive power in

the case of the 1998 crisis.

3.4. Discussion of the results

Our results indicate that ROE is associated with higher risk that materialized during the recent

crisis, and in particular systemic risk. Moreover, the results hold when we consider the 1998

crisis. Different reasons may explain this association between pre-crisis ROE and the

materialization of risk during crises, the most direct of which is a contingent explanation. As

20

argued previously, the risk management and regulatory approaches to banking activities have

fostered ROE as a main performance measure. With risk correctly assessed and capital exactly

adjusted to risk, expected ROE does not increase with bank risk. However, higher risk imply

higher realized ROE in good states of nature but results in higher losses in bad states of nature.

This contingent effect may explain our results and would apply even if risk was perfectly

assessed. In the next section, we intend to check whether the simple story whereby highly

performing firms end up doing worse in shocks applies to non-financial firms as well.

In the contingent explanation, possible mistakes in the assessment of risk and the incentives

associated with ROE as a central performance measure are not taken into account. Our second

explanation is based on mistakes. If banks fail to adequately adjust capital charges to the risk of

their assets, the contingent effect is accentuated further. Flaws in regulation and diversity in risk

management practices can explain these mistakes. Finally, as a central performance measure,

ROE provides incentives. As a target, ROE potentially creates distortions in the capital allocation

process of banks in favor of riskier assets or assets with underestimated capital charges. This

explanation may reinforce the contingent effect or constitute per se an alternative explanation for

our results. Section 5 studies the existence of incentives attached to ROE.

4. Out-of-sample tests: are banks special?

In this section we analyze the relevance of the contingent explanation for two out-of-sample

tests based non-financial firms.

21

4.1 Non-financial firms and the 2007-2008 crisis

As a first out-of-sample test, we use the same empirical design and assess the impact of

pre-crisis performance measures on the stock performance for a sample of non-financial firms

with the same size (at least $10 billion in 2006) and from the same countries. The losses suffered

by shareholders of non-financial firms were substantial, since the median firm in our sample had

a stock return of minus 41% during the crisis (compared to minus 40.7% for banks). In this shock

of comparable importance for the shareholders of banks and non-financial firms, we focus on

three different pre-crisis measures of performance (return on capital employed, return on equity

and return on assets) and test their impact on the BHR of non-financial firms during the crisis.

The results reported in regressions 1 to 3 of Table 6 indicate that ROCE, ROA and ROE

are all positively associated with the stock performance of a firm during the crisis. The results are

robust to the introduction of control variables, including beta, size, cash ratio, equity ratio, fixed

assets, short-term debt and industry dummies (two-digit SIC codes). Interestingly, and in contrast

with our results for banks, the pre-crisis beta has a strong negative impact on the stock

performance of non-financial firms during the crisis. On the contrary, investors experienced more

difficulties in properly assessing bank risk, which is consistent with the fact that banks are

relatively more opaque than other firms (Morgan, 2002).

The results from this first out-of-sample test highlight that, at odds with our results on banks,

non-financial firms with higher accounting-based performance measures before the crisis did

much better during the crisis, thereby suggesting that the contingent explanation we propose for

banks is not trivial.

22

4.2 The case of the oil industry and the 2014 oil shock

One reason for the difference between banks and non-financial firms is that the 2007-2008

crisis was a banking crisis that reflected the bad state of nature for the industry. To address this

concern, we propose a second out-of-sample test in which we use the same empirical design for a

shock that is specific to another industry. We focus on the stock performance of oil companies

after the recent collapse of crude oil prices which began in June 2014. Specifically, we test the

impact of 2013 performance measures for oil-related companies (oil extraction, petroleum

refining, etc.) on the BHR from June 20, 2014, to the end of March, 20158. We restrict the sample

to oil companies from the same countries as the banks and with total assets totalling more than

$100 million at the end of 20139. This shock is of significant importance for oil companies, since

the median firm in our sample had a stock return of minus 33% over the period. The results,

which are reported in regressions 4 to 6 of Table 6, indicate that ROCE, ROA and ROE have no

impact on the stock performance of oil companies during the oil shock when we control for

capital and asset structures, size, risk and industry (four-digit SIC code) variables. The correlation

between pre-crisis performance measures and the stock performance of oil companies in the crisis

is statistically significant, but the effect disappears once we introduce our control variables. The

results for the control variables indicate that larger firms and firms with lower beta enjoyed

higher stock performance during the oil crisis.

Overall, out-of-sample tests indicate that higher performance in good times does not result in

higher losses outside the banking industry when conditions worsen. The results suggest that the

effect we observe for banks is not trivial. Of course, they cannot rule out a contingent explanation

8 The results are unchanged if we start the calculation at the beginning of June 2014 or if we stop at the end ofDecember 2014.9 The results are unchanged if we use a $1 billion threshold.

23

in the case of banks, especially as their raw material is risk. However, they reveal a specificity of

banks, in that their main performance measure is strongly associated with risk, which is not the

case outside the banking industry. This particular feature poses a serious challenge to ROE as the

main performance measure in banks.

5. ROE and CEO pay

The contingent explanation holds even if banks adequately adjust their capital charges to risk.

Mistakes in the assessment of capital charges would tend to reinforce the contingent effect.

However, ROE is also a central performance measure in banks and as such provides incentives

for managers to game this measure. Even when capital is correctly adjusted to risk, managers can

inflate ROE in good times, by allocating capital to riskier activities. They also have incentives to

make mistakes when ROE is targeted. Heterogeneity in risk-management, flaws in regulation and

the ability to circumvent it make these mistakes possible. Moreover, disentangling “naïve” errors

from strategic behavior on the side of bank managers is not an easy task.

Anecdotal evidence suggests that ROE serves as a trigger for both bonuses and equity-based

compensation granted to bank managers10 . In a recent paper, Bennett, Gopalan and Thakor

(2015) provide strong empirical evidence that bank CEOs and executives have a larger portion of

their compensation linked to ROE than their peers from other industries. This sensitivity of bank

managers’ pay to ROE empirically confirms its role as a central performance measure in banks. It

10 For example, Reuters (4th April 2014) documents that for Goldman’s CEO, “In order to get the full award,Goldman must generate an average return on equity of at least 12 percent over a span of either three or five years, atthe board's discretion. That goal is higher than the 10 percent minimum that had been set in the prior year.”

24

also reveals that high-powered incentives to maximize ROE exist in banks. These effects are

documented on a sample of U.S. banks for the period 2006-2012. In this section, we intend to

check whether the sensitivity of bankers’ pay to ROE exist for our sample of international banks

in the years preceding the crisis. We are able to carry this analysis for a subsample of 71 banks

(276 bank-year observations) for which CEO compensation is available on BoardEx database.

In all specifications, the dependent variable is the natural logarithm of total CEO

compensation, which takes into account both cash compensation (salary plus bonus) and equity-

based compensation. We control for the size of the bank in all specifications. Talent assignment

models such as Gabaix and Landier (2008) predict that CEO pay scales with size and empirical

evidence confirms that size is a key determinant of CEO compensation, both in banks and non-

financial firms (e.g., Chemmanur, Cheng, and Zhang, 2013; Cheng, Hong, and Scheinkman,

2014). We also control for observable risk variables (beta of the stock and stock return

volatility11), which proved to be associated with CEO compensation in a sample of U.S. banks

(Cheng, Hong, and Scheinkman, 2014). In their interpretation, bank managers are compensated

when they bear extra risk. Additionally, we also control for the level of equity. Indeed, testing the

model of Berk, Stanton, and Zechner (2010), Chemmanur, Cheng, and Zhang (2013) reveal

empirically, for a sample of US non-financial firms, that CEO pay and employee wages increase

with firm leverage.

Table 7 presents the results of regressions of total CEO compensation on ROE and control

variables for the period 2002-2006. We include year dummies, to control for time-specific

variations in CEO compensation, and a deposit-taking bank dummy, to control for possible

differences in CEO pay between investment banks and deposit-taking banks. Table 7 indicates

11 The beta is computed by estimating market models with weekly returns. We use the MSCI World Index as themarket portfolio and the three-month T-bill as the risk-free rate.

25

strongly that monetary incentives associated with ROE existed in banks for CEOs in the years

leading up to the crisis. In all specifications, we observe strong sensitivity of CEO compensation

to ROE12. This provides empirical evidence that the results of Bennett, Gopalan and Thakor

(2015) hold for international banks.

Results for the control variables confirm that size is a key determinant of CEO

compensation, i.e. large banks appear to pay their CEO more than their smaller counterparts. The

negative and significant coefficient of the deposit bank dummy indicates that CEOs at investment

banks are paid more than at deposit-taking banks. We also document a strong positive association

between the equity ratio and CEO compensation. This result appears opposite to what

Chemmanur, Cheng, and Zhang (2013) find for non-financial firms. However, it has to be

interpreted by taking into account the fact that equity is adjusted to the level of risk in banks. It

therefore appears consistent with the idea that higher risk should command higher

compensation13. Nonetheless, our other risk variables (beta and volatility) have no effect on CEO

compensation.

Finally, in regression 5, we also control for a market-adjusted stock return in order to

address the possibility that banks experiencing good stock performance may at the same time

achieve higher ROE and pay their CEOs more as a result. Interestingly, the results reveal that the

market-adjusted return has no impact on the level of CEO compensation, while ROE still has a

strong impact on CEO compensation. This finding is consistent with the results from

Bhattacharyya and Purnanandam (2012) and Bennett, Gopalan and Thakor (2015) that executive

12 For the sake of place, we do not report the regressions with stock volatility, but the results are identical.13 In unreported tests, we find that density (the ratio of risk-weighted assets to total assets) is positively associatedwith CEO compensation.

26

compensation in U.S. banks depends more on short-term accounting performance metrics than on

stock return.

In unreported tests, we find that the sensitivity of bank CEO compensation to ROE is

robust to the introduction of a firm-fixed effect. Moreover, we also estimate regression 5 for each

year, in the spirit of Fama and Macbeth (1973), in which case ROE coefficients are positive and

highly significant for every year between 2002 and 2006. Overall, the results from this section

provide evidence of a strong association between total CEO compensation and ROE in the years

leading up to the crisis.

6. ROE and risk culture in banks

The results from Section 3 indicate that the association between ROE and the materialization

of risk during the crisis holds both for the recent 2007-2008 crisis and for the 1998 crisis. This

raises the interesting question of whether ROE plays a role in explaining the persistence of poor

bank performance across crises, as documented by Fahlenbrach, Prilmeier, and Stulz (2012) for

U.S. banks. The existence of an ROE culture, which Admati and Hellwig (2013) describe as a

‘deeply embedded in the banking industry’, may contribute to making some banks particularly

vulnerable to crises.

First, in Table 8, we assess whether poor performance is persistent across the two crises for

our international sample of banks, as documented by Fahlenbrach, Prilmeier, and Stulz (2012) for

U.S. banks. The results confirm that bank stock returns during the 1998 crisis have strong

predictive power for returns in the recent financial crisis. Moreover, consistent with their results,

27

unreported tests reveal that this persistence in bank performance is mainly driven by the worst

performing banks.

Next, we investigate whether ROE has any predictive power to explain membership in the

group of banks that performed worst in both crises. We explicitly investigate the attributes of this

group, running Probit regressions in which the dependent variable is a dummy equal to one if a

bank is in the bottom tercile of stock returns for both crises. Panel A of Table 9 presents the

results using the attributes of banks for 2006 and Panel B for 1997. The results strongly indicate

that ROE is a key attribute of the group of worst performers. Indeed, in both panels, ROE has

strong predictive power in explaining membership in the group of worst performing banks. Both

effects appear to be highly significant. The results for the control variables indicate that larger

banks are more likely to belong to this worst performer group. The results are robust both for

1997 and 2006, although the impact of size is lower for 1997. Consistent with Fahlenbrach,

Prilmeier, and Stulz (2012), we also find evidence that asset growth in the years leading up to the

crisis has some predictive power for membership in the bottom performer group. On the contrary,

bank capital ratio and deposit ratio have no impact on explaining membership in the group of

worst performing banks. Overall, banks with higher ROE are more likely to perform poorly in

both crises, which suggests that ROE is an important element of the risk culture described by

Fahlenbrach, Prilmeier, and Stulz (2012). Whereas previous literature emphasizes the role of

short-term finance and fast growth in making banks vulnerable, our results highlight the fact that

high ROE also accounts for this vulnerability.

7. Conclusion

28

The bank business model which emerged more than four decades ago, following risk

management and regulatory approaches to bank capital, crystallized ROE as the chief

performance metric in the banking sector. ROE is not only the main measure of bank

performance, but it also drives the allocation of resources across and inside bank divisions. In

theory, from a risk management perspective, ROE could be an appropriate performance measure

if the measurement and disclosure of risk led to a perfect adjustment of the level of bank equity.

However, extreme focalization on ROE may drive managers to take higher risk, in an industry

where complexity and opacity considerably alter the ability of outsiders to observe them

accordingly, and where bank capital regulation allows arbitrage.

Numerous observers have raised their voice to criticize the use of ROE. Jenkins14 , for

instance, begs the question: ‘But what if more fundamental change is needed to the way everyone

thinks about bank profitability? What if ROE is the wrong measure entirely?’ Admati15 goes even

further in this respect, stating that ‘ROE is a flawed and misleading measure that should not be

used to measure value creation and profitability, or to determine managerial compensation […]

The fixation on ROE in the banking sector reflects and breeds a love of leverage and risk that is

dangerous for society as a whole’.

Using financial crisis as a risk revelation event, we document a very strong positive effect of

pre-crisis ROE on the risks that materialized in the crisis. Our results clearly indicate that ROE is

an aggravating factor in making banks more vulnerable to crises. ROE also appears as a main

14 Financial Times, 7th November 2011.

15 See the article “Rethinking how banks create value,” available at:http://www.gsb.stanford.edu/news/packages/PDF/%20AdmatiFocusJune.pdf

29

variable in explaining the persistence of bank vulnerability across crises, suggesting that it

constitutes a major ingredient of risk culture.

The low return environment and the reinforcement of bank capital requirements seem to have

strengthened banks’ risk management approaches from a perspective of “economizing” equity.

Given this evolution, the role of ROE is even more important as a performance measure. From a

policy perspective, the response of the regulator, consisting of reinforcing capital requirements,

may prove to be counterproductive unless it is accompanied by a radical change in the way

performance is assessed in the banking industry and their executives paid.

30

References

Acharya, V. V., Pedersen, L. H., Philippon, T., Richardson, M. P., 2010. Measuring systemic risk.Working Paper New York University.

Acharya, V. V., Schnabl, P., Suarez, G., 2013. Securitization without risk transfer. Journal of Financial

Economics, 107(3), 515-536.

Acharya, V. V., Steffen, S., 2015. The “greatest” carry trade ever? Understanding Eurozone bank risks.Journal of Financial Economics, 115(2), 215-236.

Admati, A., Hellwig, M., 2013. The Bankers' New Clothes: What's Wrong with Banking and What to Doabout It. Princeton University Press.

Aebi, V., Sabato, G., Schmid, M., 2012. Risk management, corporate governance, and bank performancein the financial crisis. Journal of Banking & Finance, 36(12), 3213-3226.

Baele, L., De Jonghe, O., Vander Vennet, R., 2007. Does the stock market value bankdiversification? Journal of Banking & Finance, 31(7), 1999-2023.

Baker, G. P., 1992. Incentive contracts and performance measurement. Journal of Political Economy, 598-614.

Baker, G. P., Jensen, M. C., Murphy, K. J., 1988. Compensation and incentives: Practice vs.theory. Journal of Finance, 43(3), 593-616.

Balachandran, S., Kogut, B., Harnal, H., 2011. Did Executive Compensation Encourage Extreme Risk-taking in Financial Institutions? Working Paper, Columbia Business School.

Barth, J. R., Caprio, G., Levine, R., 2008. Bank regulations are changing: For better orworse. Comparative Economic Studies, 50(4), 537-563.

Beltratti, A., Paladino, G., 2013. Banks’ risk-weights and cost of equity. Bocconi University WorkingPaper, November.

Beltratti, A., Stulz, R. M., 2012. The credit crisis around the globe: Why did some banks perform better?Journal of Financial Economics, 105(1), 1-17.

Bennett, B., Gopalan, R. and Thakor, A., 2015, The structure of bankers’pay, Working Paper, WashingtonUniversity in St. Louis.

Berger, A. N., Bouwman, C. H., 2013. How does capital affect bank performance during financial crises?Journal of Financial Economics, 109(1), 146-176.

Berk, J. B., Stanton, R., Zechner, J., 2010. Human capital, bankruptcy, and capital structure. Journal of

Finance, 65(3), 891-926.

31

Bhattacharyya, S., Purnanandam, A., 2012. Risk-taking by banks: What did we know and when did weknow it. Unpublished working paper, University of Michigan.

Bolton, P., Mehran, H., Shapiro, J., 2010. Executive compensation and risk taking (No. 456). Staff Report,Federal Reserve Bank of New York.

Caprio, G., Laeven, L., Levine, R., 2007. Governance and bank valuation. Journal of FinancialIntermediation, 16(4), 584-617.

Chemmanur, T. J., Cheng, Y., Zhang, T., 2013. Human capital, capital structure, and employee pay: Anempirical analysis. Journal of Financial Economics, 110(2), 478-502.

Cheng, I. H., Hong, H., Scheinkman, J. A., 2014. Yesterday’s HeROEs: Compensation and Risk atFinancial Firms. The Journal of Finance, forthcoming.

Chesney, M., Stromberg, J., Wagner, A. F., 2011. Risk-taking incentives and losses in the financialcrisis. Swiss Finance Institute Research Paper (10-18). SSRN, 1595343.

Demirgüç-Kunt, A., Detragiache, E., Merrouche, O., 2013. Bank capital: Lessons from the financial crisis.Journal of Money, Credit and Banking, 45(6), 1147-1164.

Demirgüç-Kunt, A., Huizinga, H., 2010. Bank activity and funding strategies: The impact on risk andreturns. Journal of Financial Economics, 98(3), 626-650.

Demirgüç-Kunt, A., Huizinga, H., 2013. Are banks too big to fail or too big to save? Internationalevidence from equity prices and CDS spreads. Journal of Banking & Finance, 37(3), 875-894.

DeYoung, R., Peng, E. Y., Yan, M., 2013. Executive compensation and business policy choices at UScommercial banks. Journal of Financial and Quantitative Analysis, 48(01), 165-196.

Djankov, S., La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. The law and economics of self-dealing. Journal of Financial Economics, 88(3), 430-465.

Ellul, A., Yerramilli, V., 2013. Stronger risk controls, lower risk: Evidence from US bank holdingcompanies. The Journal of Finance, 68(5), 1757-1803.

Erkens, D. H., Hung, M., Matos, P., 2012. Corporate governance in the 2007–2008 financial crisis:Evidence from financial institutions worldwide. Journal of Corporate Finance, 18(2), 389-411.

Fahlenbrach, R., Stulz, R. M., 2011. Bank CEO incentives and the credit crisis. Journal of FinancialEconomics, 99(1), 11-26.

Fahlenbrach, R., Prilmeier, R., Stulz, R. M., 2012. This time is the same: Using bank performance in 1998to explain bank performance during the recent financial crisis. The Journal of Finance, 67(6), 2139-2185.

Fama, E. F., MacBeth, J. D., 1973. Risk, return, and equilibrium: Empirical tests. The Journal of PoliticalEconomy, 607-636.

32

Gabaix, X., Landier, A., 2008. Why has CEO Pay Increased So Much? The Quarterly Journal ofEconomics, 123(1), 49-100.

Gibbons, R., 1998. Incentives in Organizations. Journal of Economic Perspectives, 12(4), 115-132.

Kaufmann, D., Kraay, A., Mastruzzi, M., 2008. Governance matters VII: aggregate and individualgovernance indicators. World Bank Policy Research Working Paper, 4280.

Kerr, S., 1975. On the folly of rewarding A, while hoping for B. Academy of Management Journal, 18(4),769-783.

Krüger, P., Landier, A., Thesmar, D., 2014. The WACC fallacy: The real effects of using a uniquediscount rate. The Journal of Finance, forthcoming.

Laeven, M. L., Ratnovski, L., Tong, H., 2014. Bank size and systemic risk (No. 14). InternationalMonetary Fund.

Mariathasan, M., Merrouche, O., 2014. The manipulation of Basel risk-weights. Journal of FinancialIntermediation.

Mehran, H., Rosenberg, J., 2007. The effect of employee stock options on bank investment choice,borrowing, and capital (No. 305). Staff Report, Federal Reserve Bank of New York.

Minton, B. A., Taillard, J. P., Williamson, R., 2014. Financial Expertise of the Board, Risk Taking andPerformance: Evidence from Bank Holding Companies. Journal of Financial and Quantitative Analysis,forthcoming.

Modigliani, F., Miller, M. H., 1958. The cost of capital, corporation finance and the theory ofinvestment. The American Economic Review, 261-297.

Morgan, D. P., 2002. Rating banks: Risk and uncertainty in an opaque industry. American EconomicReview, 874-888.

Rajan, R. G., 2006. Has finance made the world riskier? European Financial Management, 12(4), 499-533.

Stiroh, K. J., 2004. Diversification in banking: Is noninterest income the answer? Journal of Money,Credit and Banking, 853-882.

Vallascas, F., Hagendorff, J., 2013. The Risk Sensitivity of Capital Requirements: Evidence from anInternational Sample of Large Banks. Review of Finance, 17(6), 1947-1988.

Weiß, G. N., Bostandzic, D., Neumann, S., 2014. What factors drive systemic risk during internationalfinancial crises? Journal of Banking & Finance, 41, 78-96.

33

Table 1: Summary statisticsThis table reports statistics for the whole sample of 273 banks. All variables (except BHR and MES) are computed for 2006, priorto the beginning of the crisis. BHR is the weekly buy-and-hold return from July 2007 to the end of 2008. MES is the averagestock return during the worst 5% of days of the MSCI for the period July 2007-December 2008. ROE is the ratio of pre-tax-profitto equity. ROA is the ratio of the pre-tax-profit to total assets. Tangible equity is the ratio of equity minus intangible assetsdivided by total assets. Tier 1 is the ratio of Tier 1 capital to risk-weighted assets. Deposits is the ratio of customer deposits tototal assets. Funding fragility is the ratio between the sum of deposits from banks, other deposits and short-term borrowing to totaldeposits and short-term borrowing. Non-interest income is the ratio of non-interest income to total operating income. Density isthe ratio of risk-weighted assets to total assets. Size refers to the natural logarithm of total assets. BHR 2006 is the weekly buy-and-hold return for the year 2006. Beta is the slope of the regression of weekly excess stock returns on the MSCI World excessreturn from 2004 to 2006. ADRI is the anti-director index, as revised in Djankov et al. (2008). Institution is the average of the sixfollowing indicators provided by Kaufmann et al. (2008): voice and accountability, political stability and absence of violence,government effectiveness, regulatory quality, rule of law and control of corruption. Official is an index of the power ofsupervisory authorities. Private monitoring is an index of the intensity of monitoring by the private sector. Capital is an index ofregulatory oversight of bank capital. Restrict is an index of regulatory restrictions on the activities of banks. The regulationvariables are from Caprio et al. (2007).

Number ofobservations

Mean Median Std. dev Min Max

BHR 273 -0.364 -0.407 0.323 -0.988 0.411

MES (%) 265 -3.69 -3.27 2.29 -13.26 0.001

ROE (%) 273 20.12 18.49 11.06 0.57 52.39

ROA (%) 273 1.41 1.24 1.08 0.04 9.89

Tangible equity(%)

273 6.25 5.76 4.01 -0.34 41.65

Tier 1 (%) 212 9.08 8.57 2.39 4.82 22.9

Deposits (%) 273 57.18 61.57 27.11 0.00 93.40

Funding Fragility(%)

273 24.40 12.83 26.66 0.00 100.00

Non-interestincome (%)

273 38.71 35.60 22.66 0.76 109.80

Density 190 0.60 0.59 0.17 0.15 0.92

Size 273 10.92 10.54 1.40 9.21 14.49

BHR 2006 273 0.185 0.19 0.341 -0.657 1.747

Beta 273 1.02 0.98 0.42 0.23 2.39

ADRI 273 3.82 4 0.94 1 5

Institution 273 1.13 1.24 0.50 -0.76 1.83

Official 273 11.07 12 2.15 5 14

Private monitoring 273 7.12 7 0.80 4 8

Capital 273 6.11 6 1.35 2 9

Restrict 273 9.91 11 2.11 4 15

34

Table 2: Individual risk and ROE in the 2007-2008 crisis

This table reports the OLS regressions of the buy-and-hold return to the ROE and our different control variables. In each specification, the dependent variable is the weekly buy-and-hold return fromJuly 2007 to the end of 2008. ROE is the ratio of pre-tax-profit to equity. Tangible equity is the ratio of equity minus intangible assets divided by total assets. Tier 1 is the ratio of Tier 1 capital to risk-weighted assets. Deposits is the ratio of customer deposits to total assets. Funding fragility is the ratio between the sum of deposits from banks, other deposits and short-term borrowing to totaldeposits and short-term borrowing. Non-interest income is the ratio of non-interest income to total operating income. Density is the ratio of risk-weighted assets to total assets. BHR 2006 is theweekly buy-and-hold return for the year 2006. Size refers to the natural logarithm of total assets. Beta is the slope of the regression of weekly excess stock returns on the MSCI World excess returnfrom 2004 to 2006. Country variables include ADRI, Institution, Official, Private monitoring, Capital and Restrict. Below the regression coefficient is reported the standard error in parentheses and tothe right of the regression coefficient its significance (*** significant at 1%, **significant at 5%, * significant at 10%).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

ROE -1.035*** -0.690*** -0.752*** -0.518*** -0.583** -0.761*** -0.586*** -0.665*** -0.804*** -0.705***(0.177) (0.184) (0.178) (0.196) (0.234) (0.213) (0.183) (0.225) (0.197) (0.226)

Tangible equity -0.004 -0.006 -0.005 0.028** -0.003 -0.002(0.005) (0.005) (0.005) (0.011) (0.005) (0.007)

Tier1_ratio 0.016** 0.017** 0.019**

(0.008) (0.008) (0.009)

Deposits 0.444*** 0.537*** 0.559*** 0.422*** 0.497*** 0.621*** 0.541***(0.101) (0.159) (0.159) (0.099) (0.168) (0.138) (0.164)

Funding fragility -0.295***(0.081)

Non_interest -0.398***(0.108)

Density -0.447***(0.137)

BHR 2006 -0.123* -0.095(0.069) (0.083)

Size -0.051*** -0.063*** -0.062*** -0.039** -0.031** -0.049*** -0.033** -0.039** -0.035**(0.015) (0.013) (0.015) (0.017) (0.015) (0.014) (0.015) (0.016) (0.016)

Beta 0.025 0.008 0.030 -0.058 -0.028 0.012 -0.044 0.011 -0.002(0.046) (0.048) (0.052) (0.062) (0.055) (0.047) (0.059) (0.049) (0.055)

Constant -0.801*** -0.294 0.006 -0.047 -0.202 -0.725** -0.104 -0.551 -0.316 -0.822**(0.239) (0.302) (0.275) (0.278) (0.386) (0.355) (0.329) (0.401) (0.366) (0.402)

Country variables Yes Yes Yes Yes Yes Yes Yes Yes Yes YesObservations 273 273 273 273 190 212 273 212 245 206

R-squared 0.266 0.428 0.401 0.399 0.530 0.494 0.435 0.498 0.442 0.497

35

Table 3: Individual risk and ROE quartiles for the 2007-2008 crisis

This table reports the OLS regressions of the buy-and-hold return to the ROE and our different control variables. In each specification, the dependent variableis the weekly buy-and-hold return from July 2007 to December 2008. ROE Quartile 2/Quartile 3/ Quartile 4 denote banks whose ROEs were in the second,third and fourth quartiles before the crisis, with quartile 4 corresponding to the highest ROE. Tangible equity is the ratio of equity minus intangible assetsdivided by total assets. Tier 1 is the ratio of Tier 1 capital to risk-weighted assets. Deposits is the ratio of customer deposits to total assets. Funding fragility isthe ratio between the sum of deposits from banks, other deposits and short-term borrowing to total deposits and short-term borrowing. Non-interest income isthe ratio of non-interest income to total operating income. Size refers to the natural logarithm of total assets. Beta is the slope of the regression of weeklyexcess stock returns on the MSCI World excess return from 2004 to 2006. Country variables include ADRI, Institution, Official, Private monitoring, Capitaland Restrict. Below the regression coefficient is reported the standard error in parentheses and to the right of the regression coefficient its significance (***significant at 1%, **significant at 5%, * significant at 10%).

(1) (3) (5) (2) (4) (6)ROE

Quartile 2-0.166***

(0.049)-0.184***

(0.049)-0.136***

(0.051)-0.233***

(0.054)-0.263***

(0.056)-0.231***

(0.059)

ROEQuartile 3

-0.228***(0.049)

-0.250***(0.049)

-0.173***(0.052)

-0.276***(0.057)

-0.318***(0.056)

-0.278***(0.056)

ROEQuartile 4

-0.247***(0.057)

-0.273***(0.056)

-0.196***(0.062)

-0.300***(0.063)

-0.342***(0.062)

-0.296***(0.068)

Tangibleequity

-0.004(0.005)

-0.005(0.005)

-0.005(0.005)

Tier1 ratio 0.011(0.008)

0.008(0.008)

0.012(0.008)

Deposits 0.454***(0.100)

0.469***(0.155)

Fundingfragility

-0.327***(0.083)

-0.236(0.154)

Non-interestincome

-0.358***(0.102)

-0.185(0.152)

Size -0.047***(0.015)

-0.058***(0.014)

-0.061***(0.016)

-0.030**(0.015)

-0.045***(0.014)

-0.048***(0.015)

Beta 0.024(0.046)

0.010(0.047)

0.022(0.053)

-0.046(0.052)

-0.050(0.052)

-0.061(0.057)

Constant -0.199(0.299)

0.107(0.275)

0.063(0.277)

-0.461(0.358)

-0.097(0.349)

-0.219(0.357)

Countryvariables

Yes Yes Yes Yes Yes Yes

Observations 273 273 273 212 212 212R² 0.453 0.432 0.413 0.533 0.506 0.503

36

Table 4: Systemic risk and ROE in the 2007-2008 crisis

This table reports the OLS regressions of the marginal expected short-fall (MES) to the ROE and our different control variables. In each specification, thedependent variable is the MES, defined as the average stock return of a bank over the worst 5% of days for the MSCI World for the period July 2007-December 2008. ROE is the ratio of pre-tax-profit to equity. Tangible equity is the ratio of equity minus intangible assets divided by total assets. Tier 1 is theratio of Tier 1 capital to risk-weighted assets. Deposits is the ratio of customer deposits to total assets. Funding fragility is the ratio between the sum ofdeposits from banks, other deposits and short-term borrowing to total deposits and short-term borrowing. Non-interest income is the ratio of non-interestincome to total operating income. Size refers to the natural logarithm of total assets. Beta is the slope of the regression of weekly excess stock returns on theMSCI World excess return from 2004 to 2006. Country variables include ADRI, Institution, Official, Private monitoring, Capital and Restrict. Below theregression coefficient is reported the standard error in parentheses and to the right of the regression coefficient its significance (*** significant at 1%,**significant at 5%, * significant at 10%).

(1) (2) (3) (4) (5) (6)ROE -5.069***

(1.253)-3.710**(0.196)

-5.243***(1.233)

-3.799***(1.467)

-4.472***(1.348)

-3.223*(0.234)

Tangibleequity

-0.024(0.040)

-0.037(0.039)

-0.016(0.037)

Tier1 ratio 0.076(0.059)

0.074(0.055)

0.087(0.060)

Deposits 0.783(0.760)

0.245(1.429)

Fundingfragility

-0.112(0.598)

1.993*(1.142)

Non-interestincome

-1.129(0.814)

-0.782(1.175)

Size -0.705***(0.112)

-0.604***(0.124)

-0.751***(0.102)

-0.695***(0.100)

-0.703***(0.110)

-0.589***(0.107)

Beta -0.913***(0.367)

-0.471(0.394)

-0.995***(0.357)

-0.669*(0.381)

-0.840**(0.378)

-0.444(0.384)

Constant 8.062***(2.065)

4.242(2.795)

8.829***(0.355)

3.941(2.676)

8.258***(0.348)

4.219(2.681)

Countryvariables

Yes Yes Yes Yes Yes Yes

Observations 265 205 265 205 265 205R² 0.419 0.335 0.414 0.349 0.421 0.338

37

Table 5: Individual risk and ROE for the 1998 crisis

This table reports the OLS regressions of the buy-and-hold return to the ROE and our different control variables for the 1998 crisis. In eachspecification, the dependent variable is the daily buy-and-hold return from August 3, 1998 (the first trading day of August 1998) to the day on whichthe bank attains its lowest stock price during the rest of 1998. ROE is the ratio of pre-tax-profit to equity. Tangible equity is the ratio of equity minusintangible assets divided by total assets. Deposits is the ratio of customer deposits to total assets. Funding fragility is the ratio between the sum ofdeposits from banks, other deposits and short-term borrowing to total deposits and short-term borrowing. Non-interest income is the ratio of non-interest income to total operating income. Size refers to the natural logarithm of total assets. Beta is the slope of the regression of weekly excess stockreturns on the MSCI World excess return from 1995 to 1997. Country variables include ADRI, Institution, Official, Private monitoring, Capital andRestrict. Below the regression coefficient is reported the standard error in parentheses and to the right of the regression coefficient its significance(*** significant at 1%, **significant at 5%, * significant at 10%).

(1) (2) (3) (4)

ROE-0.217*** -0.175*** -0.181*** -0.184**

(0.059) (0.055) (0.052) (0.094)

Tangible equity-1.348*** -1.374*** -1.020**

(0.493) (0.493) (0.502)

Deposits0.031

(0.051)

Funding fragility-0.034

(0.094)

Non_interest income -0.271***

(0.051)

Size-0.032*** -0.044*** -0.044*** -0.033***

(0.009) (0.010) (0.010) (0.010)

Beta-0.038** -0.030* -0.031* -0.018

(0.018) (0.017) (0.017) (0.017)

Constant -0.091 0.234 0.260 0.398

(0.223) (0.254) (0.255) (0.236)

Country variablesYes Yes Yes Yes

Observations 181 181 181 169

R² 0.353 0.394 0.393 0.474

38

Table 6: Out-of-sample tests for non-financial firms in the 2007-2008 crisis and oil companies in the 2014 oil shock.

Regressions 1 to 3 of this table report the OLS regressions of the buy-and-hold return during the 2007-2008 crisis on pre-crisis performancemeasures for non-financial firms. In the three regressions, the dependent variable is the weekly buy-and-hold return from July 2007 to the endof 2008. Regressions 4 to 6 report the OLS regressions of the buy-and-hold return during the 2014-2015 oil shock on pre-crisis performancemeasures for oil-related companies. In the three regressions, the dependent variable is the weekly buy-and-hold return from June, 2014 to theend of March 2015. ROCE, ROA and ROE, respectively, are the Return on Capital Employed, Return on Assets and Return on Equity.Equity is the ratio of equity to total assets. Cash is the ratio of cash and equivalents to total assets. Short-term debt is the ratio of short-termdebt to total assets. Fixed Assets is the ratio of net properties, plants and equipment to total assets. Size refers to the natural logarithm of totalassets. Beta is the slope of the regression of weekly excess stock returns on the MSCI World excess return from 2004 to 2006. Countryvariables include ADRI, Institution. Below the regression coefficient is reported the standard error in parentheses and to the right of theregression coefficient its significance (*** significant at 1%, **significant at 5%, * significant at 10%).

(1) (2) (3) (4) (5) (6)2007-08

crisis2007-08

crisis2007-08

crisis2014-15

Oil shock2014-15

Oil shock2014-15

Oil shock

ROCE 0.526*** 0.075(0.175) (0.166)

ROE 0.103*** 0.004(0.031) (0.210)

ROA 0.760*** 0.092(0.214) (0.084)

Equity 0.222*** 0.246** 0.171* -0.025 -0.021 -0.016(0.082) (0.093) (0.099) (0.080) (0.081) (0.082)

Cash 0.001 0.096 0.025 0.340 0.336 0.342(0.170) (0.139) (0.159) (0.216) (0.216) (0.217)

Short-term debt 0.275 0.187 0.187 0.035 0.028 0.044(0.280) (0.295) (0.289) (0.227) (0.223) (0.228)

Fixed assets -0.000 -0.000 -0.000 -0.114* -0.117* -0.112*(0.001) (0.001) (0.001) (0.063) (0.062) (0.062)

Size 0.029 0.032 0.030 0.027*** 0.026*** 0.027***(0.020) (0.021) (0.023) (0.008) (0.008) (0.008)

Beta -0.181*** -0.185*** -0.183*** -0.169*** -0.169*** -0.168***(0.044) (0.064) (0.065) (0.037) (0.037) (0.037)

Constant -0.042 0.032 -0.026 -0.117 -0.105 -0.135(0.404) (0.469) (0.501) (0.154) (0.154) (0.153)

Country variables Yes Yes Yes Yes Yes Yes

Industry dummies Yes Yes Yes Yes Yes Yes

Observations 449 449 449 338 338 338R-squared 0.379 0.369 0.377 0.605 0.605 0.606

39

Table 7: CEO compensation and ROE before the crisis (2002-2006)

This table reports the OLS regressions of the natural logarithm of total CEO compensation on ROE and control variables. In eachspecification, the dependent variable is the natural logarithm of total CEO compensation, composed of cash compensation and equity-linked compensation. ROE is the ratio of pre-tax-profit to equity. Equity is the ratio of equity to total assets. Size refers to the naturallogarithm of total assets. Beta is the slope of the regression of weekly excess stock returns on the MSCI World excess return computedannually. Market-adjusted return is the difference between the annual stock return and the market return. Deposit bank is a dummyvariable that equals one if a bank customer deposits to a total assets ratio superior to 20%. Below the regression coefficient is reportedthe standard error in parentheses and to the right of the regression coefficient its significance (*** significant at 1%, **significant at5%, * significant at 10%).

Totalcompensation

(1)

Totalcompensation

(2)

Totalcompensation

(3)

Totalcompensation

(4)

Totalcompensation

(5)ROE 2.758*** 1.81*** 2.70*** 2.76*** 2.79***

(0.622) (0.589) (0.628) (0.626) (0.657)

Size 0.14*** 0.27*** 0.29*** 0.28***(0.033) (0.039) (0.043) (0.041)

Equity 7.02*** 7.27*** 7.09***(1.669) (1.634) (1.671)

Beta -0.13(0.157)

Market-adj -0.15return (0.485)

Deposit bank -0.89*** -0.77*** -0.73*** -0.74***(0.166) (0.167) (0.172) (0.185)

Constant 8.029*** 7.45*** 5.10*** 5.03*** 8.42(0.216) (0.481) (0.664) (0.659) (10.778)

Yeardummies

Yes Yes Yes Yes Yes

Observations 276 276 276 275 275R² 0.079 0.165 0.238 0.243 0.241

40

Table 8: The persistence of bank performance across crises

This table reports the OLS regressions of the buy-and-hold return during the 2007-2008 to the buy-and-hold return of the 1998 crisis. In eachspecification, the dependent variable is the weekly buy-and-hold return from July 2007 to the end of 2008. BHR 1998 is the daily buy-and-holdreturn from August 3, 1998 (the first trading day of August 1998) to the day on which the bank attains its lowest stock price during the rest of 1998.Tangible equity is the ratio of equity minus intangible assets divided by total assets. Tier 1 is the ratio of Tier 1 capital to risk-weighted assets. Depositsis the ratio of customer deposits to total assets. Beta is the slope of the regression of weekly excess stock returns on the MSCI World excess returnfrom 2004 to 2006. Country variables include ADRI, Institution, Official, Private monitoring, Capital and Restrict. Below the regression coefficientis reported the standard error in parentheses and to the right of the regression coefficient its significance (*** significant at 1%, **significant at 5%, *significant at 10%).

(1) (2) (3)

BHR 1998 0.647***(0.144)

0.637***(0.132)

0.756***(0.144)

Tangibleequity

0.001(0.004)

Tier1 ratio 0.028***(0.009)

Deposits 0.448***(0.106)

0.456***(0.134)

Size -0.064***(0.016)

-0.041**(0.017)

-0.017(0.017)

Beta 0.014(0.054)

0.080(0.026)

0.015(0.054)

Constant -0.278(0.268)

-0.494(0.302)

-1.098***(0.356)

Countryvariables

Yes Yes Yes

Observations 230 230 184

R² 0.377 0.453 0.528

41

Table 9: Probit regressions predicting membership to the worst performer group

This table reports Probit regressions predicting whether a bank’s BHR is in the lowest tercile both in the 1998 crisis and the 2007-08 crisis. Panel Apresents results using variables in 2006 and Panel B in 1997. ROE is the ratio of pre-tax-profit to equity. Asset growth is the growth rate of totalassets over the three years preceding the crisis. Other variables are the same as in previous tables. Beta is the slope of the regression of weekly excessstock returns on the MSCI World excess return from 2004 to 2006 (1995-1997 for Panel B). Country variables include ADRI, Institution, Official,Private monitoring, Capital and Restrict. Below the regression coefficient is reported the standard error in parentheses and to the right of theregression coefficient its significance (*** significant at 1%, **significant at 5%, * significant at 10%).

PANEL A :2006

PANELB :1997

(1) (2) (3) (4) (5) (6)

ROE 4.945*** 4.592*** 4.680*** 2.297*** 1.993** 1.950**

(1.141) (1.195) (1.213) (0.645) (0.794) (0.764)

Asset growth 0.397** 0.420** 0.343 0.348

(0.201) (0.194) (0.345) (0.343)

Tangible equity -0.023 -0.011 3.501 3.185

(0.032) (0.031) (5.241) (5.379)

Deposits -0.353 0.132

(0.537) (0.632)

Funding_fragility -0.283 -0.303

(0.479) (1.035)

Size 0.232*** 0.183** 0.219** 0.181* 0.206* 0.208*

(0.085) (0.091) (0.089) (0.105) (0.121) (0.124)

Beta 0.760** 0.624** 0.728** 0.225 0.196 0.198

(0.308) (0.297) (0.304) (0.265) (0.266) (0.267)

Constant -1.391 -1.234 -1.929 1.751 0.864 0.935

(1.831) (1.917) (1.934) (2.306) (2.687) (2.688)

Country variables Yes Yes Yes Yes Yes Yes

Observations 230 228 228 181 179 179

R² 0.354 0.370 0.370 0.302 0.313 0.313

42

Appendix A: Distribution of banks by country

Country Number of banks(main sample)

Number of non-financialfirms

Australia 3 6Austria 5 0Belgium 2 4Brazil 2 7China 1 6

Denmark 4 3France 14 40

Germany 11 31Great Britain 9 33

Greece 5 2Hong Kong 7 10

India 15 0Ireland 3 2Israel 5 2Italy 12 11Japan 75 91

Malaysia 1 3Netherlands 2 9

Norway 4 2Portugal 3 2Russia 2 1

Singapore 3 3South Africa 5 1

Spain 7 9Sweden 4 6

Switzerland 13 7United States 46 159

Taiwan 10 10

TOTAL 273 460

43

Appendix B: Variable Definitions

Variable Name Definition

ADRI Anti-director index, as revised by Djankov et al.(2008).

BHR The bank’s weekly buy-and-hold return from July2007 to the end of 2008.

BHR 2006 The bank’s weekly buy-and-hold return for the year2006.

BHR 1998 The bank’s daily buy-and-hold return from August 3,1998 (the first trading day of August 1998) to the dayon which the bank attains its lowest stock price duringthe rest of 1998.

Beta Bank’s equity beta from a market model of weeklyreturns in excess of three-month T-bills from 2004 to2006, where the market is represented by the MSCIworld.

Capital Index of the regulatory oversight of bank capital, fromCaprio et al. (2007).

Density Ratio of risk-weighted assets to total assets.

Deposits Ratio of customer deposits to total assets.

Funding Fragility Sum of deposits from banks, other deposits and short-term borrowing, divided by total deposits and short-term borrowing.

Institution Average of the six following indicators fromKaufmann et al. (2008): voice and accountability,political stability and absence of violence, governmenteffectiveness, regulatory quality, rule of law andcontrol of corruption.

MES The bank’s average stock return over the worst 5% ofdays for the MSCI world index during the period July2007-December 2008.

Non-interest income Sum of net fee income, net commission income andnet trading income, divided by total operating income.

Official Index of official supervisory power, from Caprio et al.(2007).

Private monitoring Index of the private sector monitoring of banks, fromCaprio et al. (2007).

44

Restrict Index of the regulatory restrictions on bank activities,from Caprio et al. (2007).

ROA Ratio of pretax-profit to total assets.

ROE Ratio of pretax-profit to equity.

Size Natural logarithm of total assets.

Tangible equity Ratio of tangible common equity to tangible assets.

Tier 1 Ratio of Tier 1 capital to risk-weighted assets.